Multi-Purpose Room Impulse Response Dataset Measured on a 3D Spatial Grid

Introduction

The sound field inside a room depends on many factors, such as the room shape, the absorption characteristics of the materials that comprise the bounding surfaces, the furniture present in the room, and the source position and its acoustic characteristics. An increasing number of publicly available room impulse response (RIR) databases that aim to provide detailed descriptions of interior sound fields can be found in the literature. These databases can be utilized in research as well as in the development and verification of signal processing algorithms that use this information on the acoustic environment. The availability of many RIR databases covering diverse scenarios is beneficial to the community.

We provide a database of RIRs, namely the Multi-Purpose RIR (MP-RIR) dataset, which contains 68736 RIRs measured on a dense 3D grid inside a complex-shaped room. We used a measurement robot with a rotating arm that operates as a linear guide and is capable of moving a vertical, linear array of eight omni-directional microphones. Four different sources have been used and were placed at eight different positions inside the room. A detailed desciption of the measurement campaign and the dataset is presented in the paper (https://aes2.org/publications/elibrary-page/?id=22515). 

Contents of the MP-RIR dataset

In the following, the contents and the structure of the provided dataset are described:

Sk_Mrir.npy:Matrix, which contains the RIRs for all measured grid points for the loudspeaker Sk, k = 1, 2, ..., 8.The matrix has the shape [N_xy, N_z, N] = [1074 x 8 x 100096], where N_xy is the number of 2D grid positions to which the robot is moving the vertical microphone array of N_z microphones. The length of each RIR is described by N.

Mxyz.npy:Matrix, which contains the microphone coordinates of the measured RIRs and corresponds to the matrices Sk_Mrir.The matrix has the shape [N_xy, N_z, N_d] = [1074 x 8 x 3]. The indexing for the first two dimensions is the same as for the matrices Sk_Mrir, so that the microphone coordinates can be immediately retrieved for the provided RIRs. The third dimension with the length N_d gives access to the x-, y- and z-coordinate values in meters.

Setup.npzDictionary, which contains parameters related to the measurement setup, with the following keys:

angles_speakerDictionary of azimuth angles in degrees of the loudspeakers, with the keys S1, S2, ..., S8.

coord_speaker_centerDictionary, which contains the x-, y- and z-coordinates of the loudspeaker positions at the center of the base of each loudspeaker. The coordinate arrays can be accessed with the keys S1, S2, ..., S8.

coord_polygonArray of shape [4,2], which contains the x- and y-coordinates in meters of the room corners C_q, q=0,1,2,3.The first dimension of the array relates to the room corners and the second dimension relates to the coordinates. 

fsSampling rate in Hz.

T_guardGuard time in samples. The guard time provides additional samples at the beginning of the RIR to increase the quality of the RIR.

T_systemDelay of the measurement system in samples.

Further Information

The delay of the RIRs is composed of the guard time T_guard, the system delay T_system and the acoustic delay T_ac. The guard time and system delay can be retrieved from the file Setup.npz described above.

A gain alignment procedure was applied to align the output SPL between the loudspeakers, as described in the paper. Additionally, all RIRs were scaled by the same value, the maximum absolute peak of all measured RIRs. As a result, the maximum absolute value in each individual RIR is less or equal to 1.